Vibroarthrographic(VAG) signals, emitted by human knee joints, are non-stationary and multi-component in nature and time-frequency distributions(TFD) provide powerful means to analyze such signals. The objective of this paper is to classify VAG signals, generated during joint movement, into two groups(normal and patient group) using the characteristic parameters extracted by time-frequency transform, and to evaluate the classification accuracy. Noise within TFD was reduced by singular value decomposition and back-propagation neural network(BPNN) was used for classifying VAG signals. The characteristic parameters consist of the energy parameter, energy spread parameter, frequency parameter, frequency spread parameter by Wigner-Ville distribution and the amplitude of frequency distribution, the mean and the median frequency by fast Fourier transform. Totally 1408 segments(normal 1031, patient 377) were used for training and evaluating BPNN. As a result, the average value of the classification accuracy was 92.3(standard deviation ${pm}0.9$)%. The proposed method was independent of clinical information, and showed good potential for non-invasive diagnosis and monitoring of joint disorders such as osteoarthritis and chondromalacia patella.